$Q$-learning Based Network Selection Mechanism for CRNs with Secrecy Provisioning

In near future, the spectrum resources shared by primary networks to spectrum pools are likely to use unlicensed band because of its increasing popularity. Due to the open access nature, in unlicensed band communication, eavesdroppers are capable of overhearing the traffic channels. This may lead to significant throughput degradation. Existing network selection mechanisms for cognitive radio networks (CRNs) do not consider this security threat adequately. In this work, we first propose a Q-learning based throughput estimation mechanism considering the possibility of throughput degradation caused by eavesdroppers. Then, based on estimated throughput values, we formulate the network selection problem in orthogonal frequency division multiple access (OFDMA) based CRNs as an integer linear program (ILP). Next, based on the ILP formulation, we propose a greedy algorithm for network selections in OFDMA based CRNs. Finally, performance of our proposed network selection mechanism has been compared with existing approaches through system level simulations.